Directional Autonomous Torpedo Maneuver Control Using Reinforcement Learning

This paper proposes a method to optimize the autonomous torpedo maneuver path for reaching the target of torpedoes, which are explosive projectile weapons in naval operations. For flexible maneuvering of torpedoes, movement in various directions is considered. Also, the obstacles in the actual marine environment and the minimization of the waypoint that occurs when the angle of the torpedoes is changed considered to increase the efficiency of torpedo maneuvering. Consequently, this study presents the environment that reflects the action of the torpedo in various directions according to the maximum rotation angle. Torpedo maneuver strategy is formulated by applying a Markov Decision Process based reinforcement learning algorithm, Q-Learning. Compared to the general Q-Learning algorithm, the superiority of the proposed algorithm is assessed and its applicability in the actual marine environment, through the success rate of reaching the target point and the number of waypoints.